"Our data is a mess" is the most common reason given for delaying an AI project, and "we have lots of data" is the most common reason given for starting one. Neither statement tells you much. Readiness is not about how much data you hold or how tidy it feels. It is about whether the specific information a specific system needs can be used.
Start from the use case
There is no such thing as being generally ready. A support chatbot needs accurate help articles. An invoice automation needs supplier records and past invoices. A sales assistant needs clean CRM data. Decide what you want to build first, then check only the data that use case depends on. Cleaning everything before starting anything is a project that never ends.
Check 1: can a system reach it?
Information that lives in personal mailboxes, in spreadsheets on individual laptops or in a legacy application with no export cannot feed an AI system. For each data source the use case needs, find out whether there is an API, a database connection or a regular export. Access is the most frequent blocker and the least glamorous to fix.
Check 2: is it correct enough?
Take a sample of a hundred records or documents and inspect them by hand. Count the ones that are wrong, outdated, duplicated or incomplete. This takes an afternoon and gives a real figure in place of an impression.
How much error is acceptable depends on the task. A system that drafts text for review can tolerate imperfect inputs. One that acts automatically cannot.
Check 3: does anyone own it?
For each source, there should be a named person or team responsible for keeping it correct. Data without an owner decays. If the answer to "who updates this when it changes?" is silence, quality will fall after launch regardless of how well the system was built.
Check 4: are you allowed to use it?
Personal data is subject to privacy law. Customer data may be limited by contract. Some information is confidential by its nature. Establish what may be processed for this purpose, whether it may be sent to an external model provider, and where it must be stored. Raise this at the start. Finding out at the end that the legal team objects is expensive.
Check 5: is it in a usable form?
Clean text and structured records are easy to work with. Scanned documents, photographs of forms, handwritten notes and files in old formats need conversion first, and conversion introduces errors. Know which share of your material is in which state.
Check 6: is it current, and will it stay current?
An AI system built on last year's price list gives last year's prices. Check how often the source changes and how updates would reach the system. A one-time export is enough for a test. A production system needs a dependable way to stay in step with the source.
What "not ready" usually means
The result of these checks is rarely a flat no. More often it is a short list: one source needs an export built, one set of documents needs an owner and a clean-up, one question needs a legal answer. That list is the real first phase of the project, and it is far smaller than "fix all our data".
Fix only what the use case needs
Resist the wish to launch a company-wide data programme as a precondition. Improve the sources the first use case requires, deliver it, and let the next use case justify the next round of work. Each delivered system also shows people why data quality matters, which makes the following clean-up easier to get support for.
Examples and history
One more kind of data is often overlooked: past examples of the task with their correct outcome. Old tickets with their resolutions, processed invoices with their postings. These are what a system is tested against. If they exist, keep them. If they do not, start collecting now.
Summary
Readiness is specific to a use case. Check access, accuracy, ownership, permission, format and freshness for the sources that use case needs, and fix only those. A data readiness check is part of our AI consulting service. If you are unsure whether your data can support what you have in mind, we can assess it with you.